Luis von Ahn Explains How Computers and Humans Learn From Each Other

Tracy Alloway (0:02)

Bloomberg Audio Studios. Podcasts, radio, news.

Hello, and welcome to another episode of the Odd Lots Podcast. I'm Tracy Alloway.

Joe Weisenthal (0:22)

And I'm Joe Weisenthal.

Tracy Alloway (0:23)

Joe, you know, I had a life realization recently.

Joe Weisenthal (0:27)

Okay, this should be good, go on.

Tracy Alloway (0:30)

It struck me that I am spending a non-negligible amount of my time proving that I am, in fact, a human being.

Joe Weisenthal (0:39)

It's getting harder and harder.

I know what you're talking about. So we're talking, you know, you go to a website and you have to enter in a CAPTCHA, and it's like, click all these squares that has like a crosswalk on them or a truck. And like, it feels like it's just getting harder. And sometimes I'm like, no, trust me, I'm a human.

Tracy Alloway (0:55)

This is it. And every time it happens, I kind of have a moment of self doubt about whether or not, is it just me? Am I particularly bad at picking out all the motorcycles in a set of pictures? Or are they just becoming increasingly weird or perhaps increasingly sophisticated in the face of new types of technology?

Joe Weisenthal (1:16)

It's not just you. I've heard this from multiple people. And in fact, prepping for this episode, I heard people talking about exactly this.

But you know, it's like a big problem. You know, we did that WorldCoin episode. Like everyone is trying to figure out like how in a world of AI and bots and artificial intelligence, all that stuff, how do you know whether someone you're interacting with is in fact a person?

Tracy Alloway (1:38)

Yeah, and I'm glad you mentioned AI, because obviously part of this dynamic is AI seems to be getting better at solving these particular types of problems. But also they're being used more, right, to train AI models. So at this point, I think we all know why we're constantly trying to identify bikes in a bunch of photos.

But the whole idea behind CAPTCHAs is or was that humans still have an edge. So there are some things that humans are better able to do versus machines.

And one of the things that we used to talk about humans having an edge in was linguistics. So there was this idea that human language was so complex, so nuanced that machines would maybe never be able to fully appreciate all the intricacies and subtleties of the human language. But obviously since the arrival of generative AI and natural language processing, I think there's more of a question mark around that.

Joe Weisenthal (2:38)

Yeah, I mean, look, I think like a typical chat bot right now is probably better than most people at just typing out several paragraphs. It's all sort of like, seems sort of, as they say on the internet, kind of mid curve to me. It never like strikes me as like incredibly intelligent, but clearly computers can talk about as well as humans. And so it raises all sorts of interesting questions. You know, you mentioned that part of CAPTCHA is part of this like training computers, a big part of these chat bots, the so-called like real life human feedback where people say this answer is better than another, this answer is better than another as they refine the models, et cetera.

So I think there's like an interesting moment where like we're learning from computers and computers are learning from us, maybe collaboratively the two sides, you know, carbon and silicon working together.

Tracy Alloway (3:25)

I think that's a great way of putting it. Also, mid curve is such an underappreciated insult. Like calling people top of the bell curve is one of my favorite things to do online. Anyway, I am very pleased to say that today we actually have the perfect guest. We're going to be speaking to someone who was very instrumental in the development of things like CAPTCHA and someone who is doing a lot with AI, particularly in the field of linguistics and language right now. We're gonna be speaking with Luis Von Ahn.

He is of course the CEO and co-founder of Duolingo. So Luis, thank you so much for coming on Odd Lots.

Luis von Ahn (4:04)

Thank you, thank you for having me.

Tracy Alloway (4:06)

So maybe to begin with, talk to us about the idea behind CAPTCHA and why it seems to have become, I don't want to say a significant portion of my life, but I certainly spend a couple minutes every day doing at least one version.

Luis von Ahn (4:21)

Yeah, so the original CAPTCHA, the idea of a CAPTCHA was it's a test to distinguish humans from computers. The reasons why you may want to distinguish whether you're interacting with a human or a computer online, for example, and this is kind of the original motivation for it, companies offer free email services. And, you know, they have the problem that if you allow anything to sign up for a free email service, like either a computer or a human, somebody could write a program to obtain millions of free email accounts. Whereas humans, because they are usually not that patient, cannot get millions of email accounts for themselves, they can only get one or two.